The author proposes a "pragmatic" version of Functional Decision Theory (FDT) to address theoretical pitfalls identified by academics. This pragmatic FDT focuses on identifying equivalent algorithms based on demonstrable isomorphisms rather than abstract definitions, sidestepping complex logical counterfactuals. The approach involves a four-step decision process: baseline calculation, searching for likely isomorphisms, evaluating candidates, and adopting the best fit. The author also argues that when predictors make counterfactual predictions, decision theory transitions into game theory, explaining seemingly irrational choices in certain scenarios. AI
IMPACT This discussion on decision theory and game theory could influence future AI alignment research and the development of more robust AI decision-making processes.
RANK_REASON The item is an opinion piece discussing theoretical concepts in decision theory and proposing a new variant.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
